deepsurvk 0.2.1

Implementation of DeepSurv using Keras

Implementation of DeepSurv using Keras

Motivation • Features • Documentation • License • References • Credits

---

:pray: Motivation

DeepSurv is a Cox Proportional Hazards deep neural network used for modeling interactions between a patient's covariates and treatment effectiveness. It was originally proposed by Katzman et. al (2018) and implemented in Theano (using Lasagne).

Unfortunately, Theano is no longer supported. There have been some attempts in recreating DeepSurv in other DL platforms, such as czifan's DeepSurv.pytorch. However, given its popularity and ease of use, I think TensorFlow 2's Keras is a great option for this task.

mexchy1000 created DeepSurv_Keras. However, it is a very raw prototype: it is not properly documented nor validated. Moreover, it is not being actively supported anymore. Therefore, I used it as a rough starting point for the development of DeepSurvK.

This is my first Python package. I am sure there are many places where it could be improved. Feedback is always welcome!

:tada: Features

• Implemented using Keras (using TensorFlow 2)
• Includes the original datasets together with a proper description of the variables
• Designed with data as pandas DataFrames in mind
• Visualization tools for the most common plots for fast and easy exploration and prototyping
• Treatment recommender
• (Basic) hyperparameter optimization using grid and randomized search

:bookmark_tabs: Documentation

You can find the complete package's documentation here.

:page_with_curl: License

This package uses the MIT license

:black_nib: References

If you are using DeepSurvK, please cite the original DeepSurv paper, as well as the current repository as follows:

• Katzman, Jared L., et al. "DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network." BMC medical research methodology 18.1 (2018): 24. [BibTeX]
• Arturo Moncada-Torres. DeepSurvK. Accessed on [MONTH, 20XX].

:label: Credits

This package was developed in Spyder (a fantastic open-source Python IDE) using Cookiecutter and the arturomoncadatorres/cookiecutter-pypackage project template.